Give me one week of small, finished exercises over a month of ambitious intentions, every time. Most people who “start learning AI video” spend their first week the way they spent the week before: watching other people be good at it. This plan is the opposite of that.
Seven days, one narrow skill per day, one small finished clip at the end of each — sized so that a tired person with a laptop can complete them in under an hour, and sequenced so each day quietly arms you for the next.
By Sunday night you won’t be an expert. You’ll be something more useful at this stage: a person with seven posted-or-postable clips, a prompts file with honest verdicts, and hands that have touched every load-bearing part of the craft once. That’s a foundation. Everything ambitious you want to make later stands on it.
How This Week Works: Small Reps, Finished Clips
Three rules govern the week, and they matter more than any individual day’s task.
Rule one: a day ends with a finished clip, not a good one. Finished means the full loop — frame, generation, trim, one sound decision, caption, export, watched on your phone. A brilliant raw generation sitting unedited in a folder teaches you a third of what an ordinary clip carried through the whole pipeline teaches. We’re training the loop, not hunting for a masterpiece.
Rule two: one skill per day, deliberately isolated. Each day forbids the other skills from auditioning. On camera day you don’t also chase a character performance. Isolation feels artificial because it is — that’s what practice is. Musicians don’t learn scales during concerts.
Rule three: two takes per exercise, then choose. Take one shows you what the tool heard. Take two, changed by exactly one decision, shows you what your words control. A third take without a new idea is just hoping, and hope isn’t a day’s curriculum. Keep a prompts file open all week: prompt, one line of verdict, one line of what you changed. That file will be embarrassing by day four and priceless by day thirty.
Pick one generator now and refuse substitutes until the week ends — Seedance 2.5 or Google Veo both carry this curriculum easily, and the which-to-learn-first comparison can settle the choice in ten minutes. Keep durations short all week. Keep motion honest. And set up the skeleton from organising your project folders before day one if you haven’t: seven days of exercises without verdicts in filenames is just a new pile forming in real time.
Day 1: One Frame, One Motion
Today is the smallest possible unit of the craft: approve a still frame, then ask it for one gentle motion. Pick a subject with no face to lose — a steaming mug by a window, a plant in breeze-light, a boat resting on still water — or, if you must have a face, choose the gentle portrait bones from any starter prompt you trust.
The exercise: generate three stills, approve one after checking it at full size (edges, odd details, anything you wouldn’t want moving), then generate two motion takes of that frame with a single verb between them — “steam rises and drifts,” “a slow blink and a breath.” Camera still. Choose the better take, trim the dead half-second, add one quiet sound or none, caption it with what we’re looking at, export, watch on the phone.
What today installs: the frame-first reflex, the gateway habit of the whole storyboard-first workflow, and your first honest entry in the prompts file. If both takes wobble, the lesson isn’t “I am bad at this” — it’s “this frame could not afford that motion,” which is professional knowledge in a beginner’s notebook.
Day 2: Camera Moves
Subject stays boring and still today; the camera is the performer. Choose a sturdy object — a bicycle against a wall, a desk setup, a market stall — and write one slow camera instruction: a gentle push in, a slow pan across, a small arc. Generate two takes of one move only. Then, if time allows, rerun the same frame with the opposite move and compare which direction your tool renders cleaner; that asymmetry surprises everyone and it’s worth knowing about your tool early.
Keep the vocabulary plain — slow, gentle, steady — and notice that the words “cinematic” and “dynamic” do no mechanical work here. What gives a camera move its quality is one instruction, executed slowly, over a subject that isn’t also demanding motion budget. The menu of moves worth learning is short, and camera movement prompts is the reference to grow into after this week.
Today’s victory condition: one move you can predict. You’ll use that push-in for years; establishing shots are half of all openings.
Day 3: Light And Mood
Same bones as day one, new protagonist: light. Take one subject — the mug again is fine, a street corner is lovely — and generate it twice, changing only the light sentence. Morning window light, then low warm evening light. Then a third frame, if takes allow, in overcast softness. You’re building the single most underrated prompt skill a beginner can own: naming light specifically enough that the model can obey it, and watching how mood follows light around like a dog.
While the takes render, write your light vocabulary into the prompts file: direction (from the left, from behind), quality (soft, hard, diffused), time (dawn, golden hour, blue evening), and source (window, lamp, streetlight). Four slots. By day’s end you’ll have felt how “golden hour from behind” and “flat midday” aren’t adjectives but lighting rigs. The full lexicon lives in the lighting prompts guide; today is about owning four phrases with your hands. Caption today’s clip with the hour it feels like. Export the better-lit take. Notice how little else you needed.
Day 4: A Character Who Returns
The week’s boss level, placed deliberately after three days of groundwork. Create one character as a still first — clear face, neutral expression, plain background, the good reference discipline from reference images for consistent characters. Approve that portrait like it’s a cast member’s headshot, because it is. Then make two short clips of the same character doing two different small things: looking out a window, then setting down a cup; standing at a platform, then picking up a bag. Same reference, same written description pasted identically into both prompts, low motion, still camera.
Review today’s takes with the character consistency eye: not “are these identical” — they won’t be — but “would a viewer accept these as the same person in one video.” Keep whichever pair comes closest, trim both, and edit them back to back with a cut between. That tiny two-shot sequence, a character returning, is the seed of every series you’ll ever make. Log precisely what drifted and where. Day four’s notes will be the most reread page of your week, I promise you.
Day 5: Sound And Captions
Today the generator takes a back seat. Reopen your favourite clip from days one to four — the one with nothing to prove — and finish it properly, three ways. First with a matched sound bed: rain for the window shot, room tone and a distant street for the platform, kept low enough to feel rather than hear. Then with a clear caption line, checked against a phone screen and clear of the subject’s face. Then export it using your settled settings — if you haven’t settled them, the short-form export guide is today’s homework — and save the preset with your name on it like you mean it.
The lesson hiding inside day five: a mediocre generation with right sound and honest captions outperforms a pretty generation posted naked, in the only arena that counts, which is a stranger’s muted phone. Spend the session in Premiere Pro or your editor of choice and nowhere else. Boring day, enormous returns. Every future video you make will pass through this exact room; today you just furnished it.
Day 6: A Three-Shot Story
Everything converges. Write three sentences — setup, turn, land — for the smallest story you can respect: a cook alone in a lit food truck, a bus arriving, cups lined up grinning; a kite that will not rise, a gust, a run downhill laughing. One location, one character if any, one action per beat, exactly the discipline of writing a script a camera can film. Three frames as stills. Three generations, two takes allowed on the middle one only, because turns are where stories wobble. Edit in order, trim each shot to its living seconds, let the landing breathe half a second longer than feels tidy.
If your day-four character can star in it without breaking, let them — returning faces make three shots feel like a film instead of a triptych. Post this one with extra care: disclosure line if the realism asks for it, caption that names the turn, the full checklist run without hurry. This is the week’s portfolio piece, and making it will feel noticeably easier than day one felt — sit with that for a second when it happens. That ease is the week’s real product.
Day 7: Post, Review, Plan Week Two
Post the story if you held it overnight, then do the session most beginners skip and most working creators protect: the review. Spread the week’s clips out — all seven — and watch them in order on your phone, notebook open. For each, your log already has the ugly honesty; now find the patterns. Which defect kept visiting — hands, drift, pacing, light? Which prompt phrase reliably behaved? Which day’s skill surprised you by being useful everywhere?
Then plan week two out of the evidence, not out of appetite. The pattern chooses the curriculum: if faces drifted, week two is reference work and shorter turns. If edits felt clumsy, week two lives in the timeline. If every clip begged for a voice, narration earns its turn. Write five concrete sessions, each as small as this week’s days. And keep the streak’s shape — small reps, finished clips, verdicts written while you still remember why. The channel-level view of where this road leads is sketched in starting an AI video channel in 2026, and the classic failure list to keep dodging is in beginner mistakes in AI video — you’ll know several personally by now, which is how you know the week worked.
Frequently Asked Questions
What if I miss a day?
Slide the plan, don’t double up. Two exercises crammed into one tired evening teach half as much as each given its own sitting. The week is a rhythm, not a deadline — but keep the order, because the days are load-bearing in sequence.
How long should each day take?
Plan on forty-five to sixty minutes, with generation queues deciding the rest. If a day’s takes misbehave badly, finish with your better take and log the failure properly. A logged failure completes the day. An abandoned one doesn’t.
Should I post every day’s clip publicly?
Post most of them. The full loop includes an audience of strangers, however small, and day six deserves a real post with a real checklist. If a day’s result is genuinely broken, post the next best finished thing — but finish something every day. Private perfection is how weeks evaporate.
Seven days from now the plan ends, and what replaces it matters: not a bigger plan, but your own, written from your own log. You’ll know which skill is actually your bottleneck, which is a rarer and more valuable knowledge than any tutorial can hand you. Small reps, finished clips, honest notes — run that loop for a month and you’ll stop introducing yourself as someone who is “learning AI video.” You’ll just be someone who makes it, on a Tuesday, without drama. That is who this week was for.